Integrative Biomedical Research

1. Introduction

Amyotrophic lateral sclerosis and frontotemporal dementia are, in many respects, two faces of a single disease. That is not a rhetorical flourish; it is, increasingly, how the field has come to see them-two adult-onset neurodegenerative disorders that sit along one continuous clinical, pathological, and genetic spectrum rather than occupying separate diagnostic silos (Ito & Suzuki, 2011; Mackenzie et al., 2017; de La Seiglière et al., 2026). Clinically, of course, they still look quite different. ALS manifests chiefly as a relentless loss of upper and lower motor neurons, producing weakness, muscle atrophy, fasciculations, and, eventually, fatal respiratory failure (Masrori & Van Damme, 2020). FTD, by contrast, tends to announce itself through personality change and disinhibition-the behavioural variant-or through one of the primary progressive aphasia syndromes, in which language, rather than movement, unravels first (Clark, 2024; Olney et al., 2017). And yet beneath these divergent clinical presentations lies a shared, and rather striking, neuropathological signature. In more than 97% of ALS cases and roughly half of FTD cases, affected neurons and glia accumulate cytoplasmic, ubiquitinated, and hyperphosphorylated inclusions of the 43-kDa transactive response DNA-binding protein, TDP-43 (Ling et al., 2013; Neumann et al., 2006). A smaller, but no less important, subset of cases instead features inclusions of the RNA-binding protein FUS, or arises from massive hexanucleotide repeat expansions within the C9orf72 gene (Neumann et al., 2006; Renton et al., 2011; Van Langenhove et al., 2012).

For a long time-longer, arguably, than the evidence really justified-these inclusions were treated as little more than cellular debris: static, terminal deposits left behind once protein quality-control had failed (Kanekura & Kuroda, 2022). That interpretation has not aged well. A series of biophysical discoveries, beginning with the recognition that P granules behave as liquid droplets rather than solid particles, has reframed many of the ALS/FTD-linked proteins as molecules with an intrinsic, almost paradoxical, talent for demixing (Brangwynne et al., 2009; Conicella et al., 2016; Shorter, 2019). This process, liquid-liquid phase separation (LLPS), describes the spontaneous partitioning of a homogeneous macromolecular solution into two coexisting phases: a dense, concentrated, liquid-like droplet and a dilute surrounding milieu. Under ordinary physiological circumstances this is not pathology at all-it is, if anything, cell biology working as intended. LLPS underlies the formation of membraneless organelles (MLOs) such as cytoplasmic stress granules and various nuclear bodies, structures that transiently compartmentalise metabolic and signalling pathways, tune translation, and protect vulnerable transcripts during periods of cellular stress (de La Seiglière et al., 2026; Wolozin & Ivanov, 2019).

The trouble, it seems, is one of degree and duration rather than kind. The very features that make phase-separated droplets useful-their high local concentration, their close molecular apposition-also make them biophysically risky (de La Seiglière et al., 2026; Shorter, 2019). Held in that dense, crowded state for too long, or subjected to sustained cellular stress, these normally fluid, reversible droplets can begin to "mature": they stiffen, lose their liquid character, and eventually harden into gel-like or fully solid, fibrous, amyloid-like assemblies (Ishiguro et al., 2021; Kato et al., 2012; Pérez-Berlanga et al., 2023). This liquid-to-solid transition, sometimes abbreviated LST, has been proposed-not without controversy, but with mounting support-as a unifying mechanistic thread running through aggregate nucleation across the neurodegenerative disease spectrum (Alberti & Hyman, 2021; de La Seiglière et al., 2026).

What determines whether a given protein tips from reversible droplet into irreversible solid? The answer appears to lie largely in sequence. Several of the principal ALS/FTD-linked proteins-TDP-43, FUS, the dipeptide-repeat proteins generated from the C9orf72 expansion, TIA1, and UBQLN2-share a family resemblance at the level of their disordered regions, even though their folded domains and cellular functions differ substantially. Each carries stretches of intrinsic disorder, low sequence complexity, or prion-like character that predispose them toward exactly this kind of biophysical behaviour. Understanding how these sequence features, together with post-translational modifications and the surrounding cellular environment, govern the balance between reversible condensation and irreversible solidification is not merely an academic exercise. It may, in fact, be the more tractable route toward disease-modifying therapy, since intervening upstream-at the level of phase behaviour-could in principle prevent aggregation before it ever begins, rather than attempting to clear aggregates once they have already formed.

This introduction accordingly reviews the biophysical sequence determinants, the regulatory post-translational modifications, and the environmental parameters that together govern the LLPS and LST behaviour of five disease-relevant proteins: TDP-43, FUS, the C9orf72 dipeptide-repeat proteins, TIA1, and UBQLN2. Rather than treating each protein as an isolated case study, the aim here is to build toward a shared mechanistic framework-one that can accommodate the considerable differences between, say, the hydrophobic self-association of poly-GA and the electrostatically driven coacervation of arginine-rich dipeptides, while still identifying the common thermodynamic and structural rules that connect them. Where the evidence remains genuinely unsettled-and in this field, it often does-that uncertainty is noted rather than smoothed over, because the gaps in our current understanding are, arguably, as informative as what has already been established. Taken together, the sections that follow attempt to trace a coherent path from the biophysics of a single disordered domain, through the cell biology of membraneless organelles, and ultimately to the neuropathology observed at autopsy in patients with ALS and FTD.

2. Mechanisms of Phase Separation and Aberrant Liquid-to-Solid Transitions in ALS/FTD Pathobiology

2.1 The Molecular Grammar of Phase Separation: Disordered Domains as the Common Denominator

If there is a single unifying feature among the proteins implicated in ALS and FTD, it is probably this: nearly all of them carry substantial stretches of sequence that never fold into a stable tertiary structure. These intrinsically disordered regions (IDRs), sometimes further classified as low-complexity domains (LCDs) or prion-like domains (PrLDs), tend to be compositionally biased-enriched in glycine, serine, proline, and glutamine-and it is precisely this bias that allows them to engage in the kind of weak, transient, but highly multivalent interactions (hydrophobic, electrostatic, cation-π, and π-π stacking) that drive phase separation in the first place (Das et al., 2020; de La Seiglière et al., 2026; Kanekura & Kuroda, 2022; King et al., 2012; Nott et al., 2015). Figure 1 sketches the general trajectory that these domains can follow: from a soluble monomer, through weak multivalent contacts, into a liquid condensate, and-under the wrong circumstances-onward into an irreversible amyloid state. It is worth emphasising that this pathway is not obligatory; most LLPS events resolve back to the soluble state, and pathology arises only when the later, largely one-way steps are engaged (de La Seiglière et al., 2026; Shorter, 2019) (Figure 1).

2.2 TDP-43: A Transient Helix as the Fulcrum of Reversibility

TDP-43's capacity for LLPS is mediated principally by its C-terminal LCD (residues 267-414), and within that domain, a comparatively short and evolutionarily conserved stretch (residues 321-330) does a disproportionate amount of the work. NMR and simulation studies show that this segment folds, transiently and cooperatively, into a hydrophobic α-helix upon self-association, and it is this helix-together with hydrophobic and aromatic residues immediately downstream (331-340)-that forms the primary intermolecular interface stabilising the condensed phase (Conicella et al., 2016). What makes this finding clinically resonant is that several ALS-associated missense mutations-A321G, Q331K, and M337V among them-map directly onto this helical subdomain, and each one disrupts the helix-helix contacts required for reversible LLPS, tilting the equilibrium toward solid, irreversibly aggregated species instead (Conicella et al., 2016). The amyloidogenic pathway downstream of this disruption is further tuned by electrostatics: salt screens the repulsive charges that would otherwise keep the LCD soluble, thereby accelerating fibrillation (Babinchak et al., 2019; Mompeán et al., 2016). Perhaps most tellingly, the droplet environment itself appears to act as a catalyst-cross-β amyloid fibrils have been observed nucleating directly from within mature liquid droplets, rather than forming independently in bulk solution (Babinchak et al., 2019).

2.3 FUS: RNA Recognition and a Tale of Two Mutant Classes

FUS presents a somewhat more intricate picture, shaped by an extensive N-terminal prion-like domain, several arginine/glycine-rich (RGG) motifs, and a C-terminal PY-nuclear localisation signal (de La Seiglière et al., 2026; Kanekura & Kuroda, 2022). Under normal conditions, full-length FUS recognises G-quadruplex structures within specific target mRNAs (PSD-95 and CaMKIIα, for instance), and this recognition nucleates and accelerates FUS LLPS in a shape-dependent manner (Ishiguro et al., 2021). Single-molecule work has since revealed that ALS/FTD-linked FUS mutations do not behave uniformly; instead, they cluster into two biophysically distinct phenotypes (Ghanbari Niaki et al., 2020). Arginine substitutions, such as R244C, disrupt the normal dynamism of RNA binding altogether-the mutant protein essentially locks onto RNA from the moment of first contact, producing large, poorly dynamic condensates. Glycine substitutions, such as G156E, tell a different story: RNA-binding kinetics remain largely intact, yet the resulting condensates age with unusual speed, losing fluidity almost as soon as they form, which implicates glycine specifically as a flexible "spacer" residue that normally preserves condensate liquidity (Ghanbari Niaki et al., 2020). Both defects can, encouragingly, be countered pharmacologically-or at least biologically-by Karyopherin-β2 (Kapβ2), a nuclear import receptor that doubles as a molecular chaperone capable of physically dissolving aberrant FUS condensates and restoring near-wild-type RNA interactions (Ghanbari Niaki et al., 2020; Guo et al., 2018). FUS phase behaviour is additionally sensitive to oxidative modification: glutathionylation of Cys447 within the zinc-finger domain reduces FUS solubility and promotes cytoplasmic aggregation, an effect that glutathione transferase omega can, at least partially, reverse (Cha et al., 2022).

2.4 The C9orf72 Dipeptide Repeat Proteins: Two Distinct Physical Vocabularies

The GGGGCC hexanucleotide repeat expansion in C9orf72 is the single most common genetic cause across the ALS/FTD spectrum, and it drives toxicity through at least two parallel routes: autonomous RNA condensation, and repeat-associated non-ATG (RAN) translation into dipeptide repeat proteins (Fay et al., 2017; Jafarinia et al., 2020). On the RNA side, the expanded transcript folds into stable multimolecular G-quadruplexes that phase separate on their own, sequestering splicing factors into nuclear RNA foci (Alberti et al., 2025; Fay et al., 2017). On the protein side, RAN translation yields five dipeptide species, of which three-poly-GA, poly-GR, and poly-PR-carry substantial toxicity (Gendron et al., 2013; Jafarinia et al., 2020). Coarse-grained molecular dynamics modelling indicates that phase behaviour among these three is strictly length- and concentration-dependent, but-and this is the more interesting part-the underlying physical mechanism diverges sharply depending on amino acid composition (Jafarinia et al., 2020). Poly-GA, uncharged and hydrophobic, undergoes classical homotypic demixing: alanine-driven hydrophobic contacts bring chains close enough together to nucleate short-range hydrogen bonding and, eventually, aggregate formation (Jafarinia et al., 2020). Poly-GR and poly-PR, by contrast, are far too positively charged to phase separate with themselves; their arginine residues repel one another. Instead, they rely on complex coacervation with polyanionic partners-RNA, nucleolar proteins, other acidic tracts-binding stress granules and nucleoli and, in the process, disrupting nucleocytoplasmic transport (Boeynaems et al., 2017; Jafarinia et al., 2020). Figure 2 contrasts these two physical vocabularies side by side, tracing how a shared genetic lesion produces mechanistically distinct downstream pathologies depending simply on which dipeptide is being translated (Figure 2).

2.5 TIA1, UBQLN2, and the Wider Proteostasis Network

The phase-separation story does not end with the primary disease proteins; several quality-control factors modulate-and are themselves modulated by-these transitions. Rare mutations in TIA1 (P362L, A381T, E384K) strengthen homotypic interactions within this stress granule-associated LCD protein, shifting its coexistence curve leftward, delaying stress granule disassembly, and producing poorly dynamic granules that go on to recruit and insolubilise cytoplasmic TDP-43 (Mackenzie et al., 2017). UBQLN2, a proteasomal shuttle factor, undergoes LLPS through its proline-rich (Pxx) region, and ALS-linked mutations in this region (T487I, P497S) lower the concentration threshold required for phase separation while accelerating oligomerisation toward gel-like or solid states (Dao et al., 2019). Here again, however, there is a built-in rescue mechanism: ubiquitin itself preferentially binds the diffuse, monomeric form of UBQLN2, shifting the phase boundary and dissolving both wild-type and mutant aggregates (Dao et al., 2019). Collectively, these findings suggest that the proteostasis network is not a passive bystander to LLPS dysregulation but an active, if imperfect, counterweight to it.

2.6 From Reconstitution to the Neuron: Persistent Gaps and Interfacial Nucleation

For all the mechanistic clarity that in vitro reconstitution has provided, a genuine bottleneck remains in translating these findings to the intact, degenerating neuron. Cell-free assays typically rely on simplified, homotypic mixtures and artificial crowding agents, whereas the real intracellular environment is far messier-extreme macromolecular crowding, spatially restricted chaperone pools, and a dense web of competing heterotypic protein-RNA interactions that no reconstituted system fully captures (de La Seiglière et al., 2026; Kanekura & Kuroda, 2022). A further, more recent complication concerns where, exactly, amyloid nucleation begins. Emerging evidence suggests it

Figure 1. From physiological liquid-liquid phase separation (LLPS) to pathological liquid-to-solid transition (LST). Native, soluble RNA-binding proteins first engage in weak, multivalent contacts that nucleate a reversible liquid condensate (blue path), the physiological basis of membraneless organelles. Under sustained stress, post-translational modification, or disease-linked mutation, these condensates 'age' at the droplet-solvent interface into gel-like and ultimately solid, cross-β amyloid assemblies (red path), the neuropathological hallmark of ALS/FTD.

Figure 2. Divergent biophysical grammar of C9orf72-derived dipeptide-repeat proteins (DPRs). A single hexanucleotide (GGGGCC) repeat expansion, translated via repeat-associated non-ATG (RAN) translation, produces DPRs with two distinct physical logics. Poly-GA (left) is uncharged and hydrophobic, undergoing homotypic demixing above roughly fifty repeats to seed cross-β amyloid. Poly-GR and poly-PR (right) are cationic and instead undergo heterotypic complex coacervation with RNA and acidic nucleolar proteins above roughly twenty-five repeats, disrupting nucleolar and stress-granule function.

is not spatially uniform throughout a condensate but is instead catalytically favoured at the droplet-solvent interface, where the anisotropic, crowded environment appears to lower the energetic barrier to nucleation in ways the droplet interior does not (de La Seiglière et al., 2026; Linsenmeier et al., 2023; Visser et al., 2025). Exactly which cytoplasmic stress pathways, post-translational modifications, and RNA structures regulate this surface-catalysed transition in human motor neurons remains, honestly, an open and actively contested question (de La Seiglière et al., 2026)-one that the remainder of this review.

2.7 Toward Testable Questions: Rationale for the Present Synthesis

Taken together, the preceding sections point to three recurring, unresolved themes that this review-and the synthesis presented in the Results section-attempts to bring into sharper focus. First, how do differences in target RNA structure, particularly G-quadruplex conformation, differentially tune the kinetics of surface-catalysed nucleation for FUS versus TDP-43 under crowding conditions that better approximate the neuronal cytoplasm (de La Seiglière et al., 2026; Ishiguro et al., 2021)? Second, what specific thermodynamic and conformational features at the droplet-solvent interface actually decide whether a given phase-separated RNA-binding protein resolves back to the soluble state or commits irreversibly to the solid phase (Linsenmeier et al., 2023; Visser et al., 2025)? Third, to what extent do competing post-translational modifications-zinc-finger glutathionylation versus arginine methylation, for example-alter the efficiency with which chaperones such as Kapβ2 recognise and dissolve pre-formed condensates (Cha et al., 2022; Gittings et al., 2020; Guo et al., 2018)? These questions are not merely of academic interest. Each maps onto a plausible therapeutic axis-RNA-structure-targeted small molecules, interface-stabilising compounds, and chaperone-enhancing strategies, respectively, and each requires the kind of quantitative, cross-study synthesis that the Methods and Results sections below were designed to provide.

3. Methods

3.1 Overview and Rationale

Because the questions motivating this review span structural biophysics, computational modelling, and cell biology, no single experimental technique could have generated the evidence base required. We therefore adopted a structured, narrative-synthesis methodology-broadly informed by standard systematic-review reporting principles-rather than a single-technique experimental protocol. The intention, in choosing this approach, was to make the search-and-selection process reproducible enough that another investigator, working from the same terms and databases, would arrive at a substantially overlapping evidence set.

3.2 Search Strategy and Databases

We searched PubMed/MEDLINE as the primary bibliographic database, supplemented by Web of Science and Google Scholar for cross-verification, covering records from January 2006 (the year TDP-43 was first identified as the pathological substrate common to ALS and FTD; Neumann et al., 2006) through the most recent indexed literature. Search strings combined disease terms ("amyotrophic lateral sclerosis" OR "ALS" OR "frontotemporal dementia" OR "FTD" OR "C9orf72") with biophysical process terms ("liquid-liquid phase separation" OR "LLPS" OR "biomolecular condensate" OR "membraneless organelle" OR "liquid-to-solid transition" OR "amyloid nucleation") and, where relevant, protein-specific terms ("TDP-43" OR "FUS" OR "dipeptide repeat" OR "TIA1" OR "UBQLN2" OR "MATR3"), combined using standard Boolean operators. MeSH terms were mapped where available to maximise sensitivity.

3.3 Eligibility Criteria

Included sources were peer-reviewed primary research articles, methodologically detailed reviews, and structural or biophysical studies published in English that directly addressed the phase behaviour of ALS/FTD-associated proteins or nucleic acids, whether investigated in vitro (recombinant protein reconstitution), in silico (coarse-grained or atomistic molecular dynamics), or in cellulo (patient-derived or model cell systems). Studies were excluded if they addressed unrelated protein aggregation phenomena without a stated or reasonably inferable connection to ALS/FTD biology, if they were conference abstracts lacking sufficient methodological detail, or if full text could not be retrieved through institutional or open-access channels.

3.4 Study Selection and Data Extraction

Titles and abstracts were first screened for topical relevance, after which full texts of potentially eligible articles were retrieved and assessed against the eligibility criteria above. From each included study, we extracted, where reported: the specific protein or nucleic acid construct under study; experimental or simulation conditions (pH, ionic strength, temperature, crowding agent, and concentration); quantitative phase-behaviour parameters (saturation concentration, critical repeat length, polymer scaling exponent, persistence length, dilute- and dense-phase concentrations); kinetic parameters (fibrillation lag time, droplet fusion rate, fluorescence recovery after photobleaching half-time); and any reported chaperone, post-translational modification, or small-molecule modulator of the phase transition. Extracted data were tabulated (Tables 1-4) to allow direct, side-by-side comparison across constructs and studies, a step that also served as an internal consistency check-values that appeared discordant across sources were flagged and re-verified against the original publication.

3.5 Synthesis Approach

Given the biophysical and methodological heterogeneity across the included studies-ranging from single-molecule spectroscopy to cell-based imaging-a formal meta-analytic pooling of effect sizes was neither appropriate nor attempted. Instead, we performed a structured narrative synthesis, organised around the shared mechanistic themes identified in the Literature Review (sequence determinants, electrostatic tuning, RNA cooperativity, and chaperone-mediated reversal), with quantitative values reported descriptively and cross-referenced to their source study throughout the Results section. Where multiple independent studies reported convergent findings for the same construct (for example, the critical repeat length for poly-GA phase separation), this convergence is explicitly noted as a form of internal validation.

3.6 Reproducibility Considerations

To support reproducibility in line with PubMed-indexed reporting norms, we have specified, for every quantitative parameter reported in the Results and Tables, the originating study and, where available, the specific experimental or computational condition under which it was obtained. Readers seeking to replicate the coarse-grained molecular dynamics findings summarised here are directed to the one-bead-per-amino-acid (1BPA) simulation framework described by Jafarinia et al. (2020); readers seeking to replicate the NMR and turbidity-based reconstitution findings for TDP-43 are directed to the protocols described by Conicella et al. (2016) and Babinchak et al. (2019). No new primary experimental or simulation data were generated for this synthesis; all reported values are drawn directly from, and attributable to, the cited primary literature.

4. The Biophysical Grammar of Pathological Phase Transitions in ALS/FTD

4.1 Overview

Pulling together the quantitative parameters extracted across the included studies, a reasonably coherent picture emerges of how physiological LLPS tips into pathological, solid-state aggregation across the five protein systems considered here. What follows synthesises those parameters-condensed into Tables 1 through 4 and Figures 3 and 4-while trying not to lose sight of the biological story each number is actually telling (de La Seiglière et al., 2026; Shorter, 2019) (Table 1; Table 2; Table 3; Table 4).

4.2 Polymer Scaling Laws and Coacervation Thresholds of RAN-Translated Dipeptide Repeat Proteins

Coarse-grained molecular dynamics, calibrated against experimentally derived polymer behaviour, indicates that the single-molecule conformation of each dipeptide repeat protein (DPR) tracks quite closely with its amino acid composition (Jafarinia et al., 2020). At matched chain length, the three toxic DPRs order themselves, in terms of spatial extension, as poly-PR > poly-GR > poly-GA-a ranking driven largely by electrostatic self-repulsion in the arginine-rich species and by hydrophobic collapse in poly-GA (Jafarinia et al., 2020). Quantitatively, poly-PR displays a polymer scaling exponent of ν = 0.70 ± 0.02 and poly-GR of ν = 0.67 ± 0.02, both above the ν ≈ 0.60 expected for a standard excluded-volume chain, whereas poly-GA sits at ν = 0.48 ± 0.02, close to the theta-solvent value expected of a compact random coil (Hofmann et al., 2012; Jafarinia et al., 2020) (Figure 3; Table 1).

Notably, this dimensional difference between poly-PR and poly-GR cannot be attributed to electrostatics alone, since proline is, if anything, the more hydrophobic of the two residues being substituted. Simulations that computationally swap the hydrophobicities of glycine and proline while holding charge constant show that it is backbone stiffness, not chemistry per se, that dominates: proline's cyclic side chain imposes a persistence length of 5.04 bond lengths, versus 2.06 bond lengths for the far more flexible glycine-containing poly-GR (Jafarinia et al., 2020).

These conformational differences translate directly into distinct phase-separation thresholds. Poly-GA, being uncharged, undergoes purely homotypic LLPS driven by long-range hydrophobic attraction, but only once repeat length exceeds n = 50; below that cutoff, no amount of concentration appears sufficient to trigger demixing (Jafarinia et al., 2020). Within the 60 ≤ n ≤ 100 range, the dilute-phase concentration spans roughly 6-33 mg/mL while the dense phase reaches an extraordinary 500-670 mg/mL-densities high enough, evidently, to bring GA chains into the kind of close proximity needed to nucleate short-range hydrogen bonding and, from there, cross-β amyloid structure (Jafarinia et al., 2020) (Table 1). This n = 50 threshold sits remarkably close to the experimentally observed range of 46-61 repeats required for cytoplasmic poly-GA aggregation in Neuro2a cells, which is reassuring, in the sense that a purely computational prediction converges so tightly with cell-based data (Jafarinia et al., 2020).

Poly-GR and poly-PR, by contrast, simply cannot phase separate with themselves-their mutual electrostatic repulsion is too strong-so instead they rely on heterotypic complex coacervation with polyanionic partners such as RNA or acidic nucleolar proteins (Jafarinia et al., 2020). For poly-PR interacting with a physiologically representative polyaspartate model (D40), the minimum repeat length required to trigger coacervation is n = 25, at an optimal stoichiometry of rPR = 0.62 (Jafarinia et al., 2020). This threshold aligns closely with the roughly 30-repeat clinical toxicity boundary observed in patient-derived cells, suggesting that the acquisition of coacervation capacity is itself a meaningful step in pathogenesis, not merely a laboratory curiosity (Jafarinia et al., 2020) (Figure 3; Table 1). Because poly-GR's more flexible backbone allows it to conform more readily around anionic partners, it binds acidic targets like nucleolin and NPM1 more tightly than the stiffer poly-PR, which in turn helps explain why poly-GR appears especially effective at displacing native nucleolar proteins and driving nucleolar stress in patient cells (Jafarinia et al., 2020; Lee et al., 2016; White et al., 2019).

4.3 Structural Elements and Electrostatic Control of TDP-43 Phase Transitions

TDP-43-the protein found in cytoplasmic inclusions across the large majority of ALS cases and roughly half of FTD cases-relies for its LLPS behaviour on a short, conserved, and otherwise unremarkable-looking stretch of its C-terminal LCD (residues 319-341) that folds transiently into an α-helix populating around 50% of the monomeric structural ensemble (Babinchak et al., 2019; Conicella et al., 2016; de La Seiglière et al., 2026) (Table 2). This helix functions, in effect, as a molecular "sticker": initial contact between two helices stabilises and extends the folded structure across residues 321-340, and it is this folding-upon-assembly behaviour that underpins physiological LLPS (Conicella et al., 2016).

The mutational landscape mapped onto this helix is informative precisely because it segregates into distinct functional classes. Destabilising mutations-A321G and Q331K-break the cooperative helix outright and abolish LLPS entirely in vitro; M337V reduces salt-dependent turbidity by roughly half (Conicella et al., 2016). Engineered helix-breaking substitutions, A326P and M337P, likewise eliminate phase separation, even though M337P still permits monomeric helicity, which tells us that helix formation alone is not sufficient-the specific hydrophobic contact interface matters too (Conicella et al., 2016). Perhaps most instructive, though, is A321V, which increases local hydrophobicity, strengthens the initial LLPS driving force, and yet accelerates the transition toward irregular, solid-state, detergent-insoluble aggregates that fail to fuse-a reminder that more phase separation is not necessarily protective, and can in fact be actively harmful (Conicella et al., 2016) (Table 2).

Electrostatics exert a second, largely independent layer of control. At acidic pH the LCD carries substantial net positive charge and remains monomeric and soluble; as pH rises toward physiological values, or as ionic strength increases, this repulsive charge is screened, permitting the hydrophobic helices to associate and condense (Babinchak et al., 2019). Once formed, these droplets behave as thermodynamic accelerators of amyloid formation: under non-LLPS conditions the LCD fibrillates slowly, with a lag phase of 10-15 hours, but under LLPS-permissive conditions this lag collapses to roughly 1-3 hours (Babinchak et al., 2019) (Figure 4; Table 2). Consistent with the interfacial-nucleation hypothesis introduced in the Literature Review, high-resolution imaging indicates that fibrils preferentially emerge from the droplet-solvent interface rather than distributing uniformly throughout the dense phase (de La Seiglière et al., 2026).

Table 1. Comparative biophysical and polymer-scaling properties of ALS/FTD-related C9orf72 dipeptide-repeat proteins (DPRs). This table summarises the charge state, critical repeat length required for phase separation, polymer scaling exponent (ν), backbone persistence length, and primary cellular target for each dipeptide-repeat species and comparator polymer discussed in the Results (Section 4.2). Values are drawn from coarse-grained one-bead-per-amino-acid (1BPA) molecular dynamics simulations and complementary biophysical studies (Hofmann et al., 2012; Jafarinia et al., 2020; Ryan et al., 2018). Rows are grouped by mechanism: homotypic hydrophobic demixing (poly-GA), heterotypic electrostatic coacervation (poly-PR, poly-GR and their polyanion complexes), non-phase-separating comparators (poly-GP, poly-PA), and theoretical/experimental controls.

Table 2. Thermodynamic, kinetic, and structural parameters of the TDP-43 low-complexity domain (LCD) across environmental conditions and ALS-associated mutations. This table compares hydrodynamic radius/oligomeric state, phase-separation propensity, and fibrillation lag time for wild-type TDP-43 constructs across pH, salt, and crowding conditions, alongside representative ALS-associated and engineered mutants discussed in Section 4.3. Data are drawn from turbidity assays, dynamic light scattering, EPR spin-labelling, and solution NMR (Babinchak et al., 2019; Conicella et al., 2016). Shorter fibrillation lag times indicate accelerated amyloid nucleation under LLPS-permissive conditions.

4.4 RNA Cooperativity, Mutational Mechanics, and Chaperone-Mediated Disassembly

FUS-RNA interactions display a length-dependent stoichiometry, with monomeric FUS engaging single-stranded RNA through a bipartite, two-step binding mode, and FUS multimers forming small, fluid, reversible ribonucleoprotein coacervates with longer transcripts (Ghanbari Niaki et al., 2020) (Table 3). ALS-linked mutations split cleanly into two mechanistic subclasses. Arginine mutants (R216C, R244C, R514G, R521C/G) abolish the dynamic, multimer-dependent RNA-binding mode, locking the protein into static RNA contacts that seed large, poorly dynamic condensates. Glycine mutants (G156E, G187S, G225V, G230C, G399V) preserve near-normal RNA-binding kinetics but lose backbone flexibility, accelerating gelation and fluidity loss within minutes of initial phase separation (Ghanbari Niaki et al., 2020) (Table 3). G-quadruplex RNA binding further accelerates FUS's liquid-to-solid transition, generating rigid, hexanediol-resistant aggregates, and clinically relevant mutations divide again into loss-of-function (diminished G4 binding; P18S, R521C) and gain-of-toxicity (unregulated aggregation; R383C, P525L) categories (Ishiguro et al., 2021).

UBQLN2 tells a broadly parallel story through its proline-rich (Pxx) domain. ALS-linked substitutions there (T487I, P497L/H/S, P506S/T/A) shift the native monomer-oligomer equilibrium toward higher-order assemblies even at low micromolar concentrations, lowering the saturation concentration required for LLPS and producing viscoelastic, solid-like condensates with markedly reduced fusion kinetics (Dao et al., 2019) (Table 3). Encouragingly, two independent rescue mechanisms have been identified across these systems: ubiquitin, added at a simple 1:1 stoichiometry, completely disassembles both wild-type and mutant UBQLN2 condensates by disrupting UBA-Pxx multivalency (Dao et al., 2018), while Karyopherin-β2 performs an analogous role for FUS, binding the PY-NLS domain to dissolve condensates and restore near-native RNA interaction dynamics (Ghanbari Niaki et al., 2020; Guo et al., 2018). MATR3, a more recently characterised intrinsically disordered RNA-binding protein, forms distinct nanoscale spherical and wormlike assemblies whose reversibility is likewise governed by multivalent and RNA-dependent interactions, with disease mutations shifting these assemblies toward less reversible states (Sprunger et al., 2025) (Table 3).

4.5 Native Cellular Oligomers and Stress-Dependent Divergence Between FUS and TDP-43

In vitro reconstitution necessarily simplifies a considerably messier cellular reality, and single-molecule pulldown experiments performed directly in neuroblastoma cells help close that gap. Wild-type FUS exists predominantly as a monomer or dimer under resting conditions, but ALS-linked arginine and glycine mutants show elevated basal oligomerisation, with the most severe NLS mutants (R495X, P525L) forming stable four-to-six-subunit multimers that resist dissolution by high salt, hexanediol, RNase A, or even Karyopherin-β2, yielding only to harsh denaturants such as 8 M guanidine hydrochloride (Table 3). This resistance suggests these are not ordinary liquid condensates at all, but pre-formed, amyloid-like nuclei already present in the resting cytoplasm.

Hyperosmotic stress reveals a further, and arguably more clinically relevant, divergence between FUS and TDP-43. Both proteins undergo reversible nuclear LLPS in response to acute crowding, but their downstream trajectories differ sharply: FUS droplets recruit an active Hsp70/Hdj2/DnaJC7 chaperone network that maintains liquidity and permits rapid, complete disassembly once stress resolves, whereas TDP-43 droplets fail to recruit this same chaperone machinery and, under sustained stress, undergo an essentially irreversible liquid-to-solid transition (Mackenzie et al., 2017) (Table 4). This asymmetry-one protein protected by an active chaperone response, the other left comparatively exposed-may go some way toward explaining why TDP-43 nuclear depletion and cytoplasmic mislocalisation are such consistent features of ALS/FTD neuropathology, even though FUS is subject to broadly similar stress exposure in the same cells (Mackenzie et al., 2017) (Table 4).

5. Reconciling Divergent Biophysical Grammars into a Shared Therapeutic Roadmap

5.1 A Common Thermodynamic Logic Beneath Superficially Different Proteins

Stepping back from the individual protein systems, what stands out most is not their differences-considerable as those are-but the underlying logic they seem to share. TDP-43, FUS, the C9orf72 dipeptides, TIA1, and UBQLN2 each rely on weak, multivalent, low-affinity interactions to establish a reversible condensed phase, and in each case, disease-associated mutations act by tipping the balance of that same thermodynamic equilibrium, whether toward

Table 3. Biophysical profiles and phase behaviour of FUS, UBQLN2, and MATR3 wild-type and disease-linked mutant systems. This table compares mutation site, saturation concentration, condensate fluidity, and the chaperone or ligand capable of rescuing aberrant phase behaviour for FUS, UBQLN2, and MATR3 wild-type and mutant constructs discussed in Section 4.4 (Dao et al., 2019; Ghanbari Niaki et al., 2020; Sprunger et al., 2025). Rescue modulators indicate the physiological factor shown experimentally to dissolve or reverse pathological condensates for each system.

Table 4. Cellular dynamics, stress responsiveness, and condensation profiles of FUS, TDP-43, TIA1, and C9orf72 repeat RNA under physiological and pathological stress. This table synthesises in vivo assembly thresholds, disassembly kinetics, chaperone recruitment, and cellular-viability outcomes for endogenous FUS, TDP-43, TIA1, and repeat RNA condensates under hyperosmotic, oxidative, and repeat-expansion stress, as discussed in Section 4.5 (Fay et al., 2017; Gao et al., 2022; Mackenzie et al., 2017). Comparisons highlight the differential chaperone recruitment that distinguishes reversible FUS condensates from the comparatively chaperone-deficient and stress-vulnerable TDP-43 condensates.

premature solidification (TDP-43 A321V; Conicella et al., 2016), toward static, non-dynamic binding (FUS arginine mutants; Ghanbari Niaki et al., 2020), or toward lowered saturation thresholds and accelerated oligomerisation (UBQLN2 Pxx mutants; Dao et al., 2019) (Table 2; Table 3). This convergence is, arguably, the single most therapeutically useful finding to emerge from the biophysics literature over the past decade, because it implies that interventions need not be bespoke to each protein-strategies that stabilise the liquid state, or that restore chaperone-mediated dissolution, could plausibly generalise across the ALS/FTD spectrum rather than requiring a separate drug for every genetic subtype.

5.2 The Interface as an Emerging, and Perhaps Underappreciated, Drug Target

One theme that recurred across several of the systems reviewed here-TDP-43 in particular, but plausibly FUS as well-is that amyloid nucleation does not occur uniformly throughout a condensate; it is concentrated at the droplet-solvent interface (de La Seiglière et al., 2026; Linsenmeier et al., 2023; Visser et al., 2025). If that is correct, and the evidence assembled in the Results section (Figure 4; Table 2) is reasonably consistent with it, then the interface itself becomes a distinct and somewhat underexplored pharmacological target-conceptually separable from either the soluble monomer or the mature aggregate, which are the two states most drug discovery programmes have historically focused on. Small molecules or peptides designed to alter interfacial tension, hydration, or curvature might, in principle, suppress nucleation without necessarily dissolving the condensate altogether, which could be an important distinction given that complete dissolution of physiological membraneless organelles is presumably not a therapeutically desirable outcome in its own right.

5.3 Chaperones and Post-Translational Modification as Two Sides of the Same Rescue Strategy

Across nearly every system reviewed, a rescue mechanism of some kind has been identified-ubiquitin for UBQLN2, Karyopherin-β2 for FUS, glutathione transferase omega for oxidatively modified FUS, and arginine methylation for the arginine-rich DPRs (Cha et al., 2022; Dao et al., 2019; Gittings et al., 2020; Guo et al., 2018) (Table 3; Table 4). What is striking, on reflection, is how these mechanisms cluster into really just two conceptual categories: charge neutralisation (methylation, glutathionylation reversal) and chaperone-mediated multivalency disruption (ubiquitin, Karyopherin-β2). This is encouraging from a drug-development standpoint, since it suggests that a comparatively small number of pharmacological strategies-enhancing PRMT activity in a targeted fashion, or boosting Kapβ2 expression or activity in vulnerable motor neurons-might address multiple disease-relevant proteins simultaneously, rather than requiring an entirely separate mechanism for each one.

5.4 Limitations, Persistent Uncertainties, and the Reconstitution-to-Neuron Gap

None of this should be read as more settled than it is. Much of the quantitative data synthesised here-critical repeat lengths, saturation concentrations, scaling exponents-derives from cell-free reconstitution or coarse-grained simulation, and the single-molecule pulldown data reviewed in Section 4.5 make clear that native cellular oligomers can behave quite differently from anything observed in a test tube (Table 3; Table 4). It remains genuinely unclear, for instance, whether the interfacial nucleation mechanism proposed for TDP-43 in vitro operates identically within the crowded, chaperone-rich, and spatially compartmentalised environment of an actual motor neuron, and it is entirely possible that additional, as-yet-uncharacterised cellular factors modulate this process in ways the current literature simply has not captured (de La Seiglière et al., 2026). Similarly, most of the DPR coacervation data reviewed here (Table 1) rely on simplified polyanionic model partners such as polyaspartate, which, while a reasonable approximation of acidic nucleolar tracts, is unlikely to fully recapitulate the structural specificity of genuine interactions with nucleolin or NPM1.

5.5 Toward the Next Generation of Mechanistic and Therapeutic Studies

Given these gaps, the most productive next steps probably lie in bridging exactly the divide identified above: applying the super-resolution and single-molecule imaging approaches that have already proven informative for FUS oligomer detection (Table 3) to TDP-43 and the C9orf72 dipeptides directly within iPSC-derived or patient motor neurons, ideally under conditions that combine physiological crowding with real cellular stressors rather than artificial PEG-based surrogates. Equally, testing whether Kapβ2 overexpression or PRMT-targeted methylation strategies can rescue phase behaviour

Figure 3. Length- and sequence-dependent phase behaviour of C9orf72 dipeptide-repeat proteins. (A) Critical repeat length required for phase separation is markedly lower for the heterotypically coacervating, arginine-rich species (poly-PR, poly-GR) than for the homotypically demixing poly-GA. (B) Polymer scaling exponents (ν) show that arginine-rich DPRs adopt extended chain conformations (ν > 0.60) driven by electrostatic self-repulsion, whereas poly-GA remains compact (ν ≈ 0.48), consistent with hydrophobic collapse. Values summarised from Jafarinia et al. (2020) .

Figure 4. Mutation-dependent shifts in TDP-43 low-complexity domain (LCD) fibrillation kinetics under LLPS-permissive conditions. Wild-type TDP-43 CTD fibrillates rapidly once phase-separated; the hyper-aggregating A321V and helix-breaking A326P variants further shorten the lag phase, while the engineered M337P variant, which retains monomeric helicity but cannot self-associate, markedly delays fibrillation. Values summarised qualitatively from Babinchak et al. (2019) and Conicella et al. (2016).

specifically in a neuronal, rather than purely biochemical, context would represent a meaningful and fairly direct translational step forward from the biophysical groundwork reviewed here.

6. Conclusion

Taken as a whole, the evidence synthesised in this review supports liquid-liquid phase separation as a genuinely unifying-if still incompletely understood-mechanism connecting the genetics, cell biology, and neuropathology of ALS and FTD. Rather than arising directly from folded, native proteins, the pathological inclusions that define this disease spectrum appear to pass through an intermediate, biophysically reversible condensed state before committing to solid, amyloid-like structure. What differs across proteins is not whether this general sequence occurs, but the specific molecular grammar governing it: TDP-43 depends on a transient α-helical "sticker," FUS bifurcates according to arginine- versus glycine-substitution class, and the C9orf72 dipeptides split further still into hydrophobic homotypic demixing (poly-GA) and electrostatic heterotypic coacervation (poly-GR, poly-PR). Despite this diversity, recurring themes-interfacial nucleation, chaperone-mediated reversal, and charge-dependent regulation-suggest that a relatively focused set of therapeutic strategies could, in principle, address multiple disease-relevant proteins at once. The persistent gap between cell-free reconstitution and the intact neuron remains the field's most pressing limitation, and closing it, through native-context imaging and physiologically realistic stress paradigms, represents the clearest path toward translating this biophysical framework into disease-modifying therapy for ALS and FTD.

References